Advanced Techniques in Urban Remote Sensing

Transcription

1 Advanced Techniques in Urban Remote Sensing Manfred Ehlers Institute for Geoinformatics and Remote Sensing (IGF) University of Osnabrueck, Germany

2 Contents Urban Remote Sensing: High Spatial Resolution Image Fusion: Iconic Image Fusion: Decision Based Change Detection Proposed SPP Conclusions

3 Taxonomy of Remote Sensing Systems Taxonomy of Remote Sensing Systems Recording Platform Recording Mode Recording Medium Spectral Coverage Spectral Resolution Radiometr. Resolution Spatial Ground Resolution Satellite/Shuttle Aircraft/Balloon Stationary Passive (Electrooptical, Thermal Infrared, Active (Laser, Radar) Thermal Microwave) Analog (Camera, Video) Digital (Whiskbroom, Line Array, 2D CCD) Visible/Ultraviolet Reflected Infrared Thermal Infrared Microwave Panchromatic 1 Band Low (< 6 bit) Very Low > 250 m Low m Multispectral 2 20 Bands Medium (6 8 bit) Medium m Hyperspectral Bands High (8-12 bit) High 4 10 m Very High 1 4 m Ultraspectral > 250 Bands Very High (> 12 bit) Ultra High < 1m

4 Image Fusion: Why is it Necessary? Remote sensors have different spatial resolution for panchromatic and multispectral imagery The ratios vary between 1:2 and 1:8 Some sensors have only 1 pan band (WorldView 1, EROS B) Some sensors have only multispectral bands (RapidEye) For multisensor fusion the ratios can exceed 1:20 (e.g. WorldView/SPOT)

5 Image Fusion Levels Source: Pohl, C. and Genderen, J.L. van, 1998, Multisensor image fusion in remote sensing: concepts, methods and applications. Int. J. Remote Sensing, Vol. 19, pp

6 Multisensor Fusion Case Study: Datasets Santo Domingo de la Calzada, Spain (SDOM) Satellite Sensor Recording Date Ground Sampling Distance (GSD) Ikonos II 30 May m (pan only) SPOT 4 3 November m SPOT 4 24 April m SPOT 4 15 May m SPOT 5 24 July m (20 m SWIR) SPOT 2 10 December m SPOT 5 10 April m (20 m SWIR) SPOT 4 20 July m Formosat 2 12 August m

7 Original Data: SPOT 5 Sample Original multispectral Spot 5 image, acquisition date: 20 July 2005 (resampled to 1 m resolution) Original Ikonos panchromatic image, acquisition date: 3 May 2005

8 Fusion Results: Brovey Original multispectral Spot 5 image, acquisition date: 20 July 2005 Fused with Brovey

9 Modified IHS Original multispectral Spot 5 image, acquisition date: 20 July 2005 Fused with Modified IHS

10 Principal Component Original multispectral Spot 5 image, acquisition date: 20 July 2005 Fused with Principal Component

11 Additive Wavelet Intensity Proportional (AWLP) Original multispectral Spot 5 image, acquisition date: 20 July 2005 Fused with AWLP

12 Ehlers Fusion Original multispectral Spot 5 image, acquisition date: 20 July 2005 Fused with Ehlers Source: Ehlers, M., S. Klonus, P. Astrand and P. Rosso, Multi-Sensor Image Fusion for Pansharpening in Remote Sensing, International Journal for Image and Data Fusion (IJIDF), Vol.1, No. 1, pp

13 Concept: Ehlers Fusion Basis: IHS Transform and Filtering in the Fourier Domain Multispectral Image Panchromatic Image FFT Fourier Spectrum HPF Pan HP FFT -1 R G B I H S FFT Fourier Spectrum LPF I LP I LP +Pan HP H S R G B IHS -1

14 Case Study: Decision Based Fusion Problem: Extent of New Settlement Areas Fast & Accurate Update Satellite Based Method High Accuracy (> 90%) Low Cost Statewide Inventory Possible Pixel Based Fusion not Suitable

15 Results: Test Site Test area near the city of Aachen (Size: 25 km 2) KOMPSAT (panchromatic) ASTER (multispectral)

16 Basic Principle: Decision Based Fusion

17 Methodology Methodology: Increasing threshold values with each lower level of the segmentation network Classification attributes for class settlement : Texture parameter (GLC-Matrices) Shape parameter (degree of compactness, length) NDVI (elimination of vegetated areas) Relations within the network (determination of nonsettlement areas)

18 Level 0: Possible settlement areas (red) Results

19 Level 1: Possible settlement areas Results

20 Level 2 + fine tuning: Settlement areas End Results

21 Overall Accuracy Level: Users accuracy: Level % Level % Level % End level %

22 Change Analysis Catastrophe Archive Reference image Action?!!?!? Automated change detection and analysis

23 CEST: Decision Tree

24 Quickbird Subsets of Abu Suruj Quickbird panchromatic image: March 2, 2006 (T1) Quickbird panchromatic image: February 28, 2008 (T2) Manually digitized change image

25 Result: Image Difference

26 Result: PCA

27 Result: Post Classification

28 Result: CEST

29 Accuracy Assessment: Coefficient

30 Rapid Change Map

31 Adding GIS Information Before Vector event map Attribute information Selected objects Post event Change detection Objects geometry and spatial location 78

32 Methodology => Information about contour integrity Detected Part of Contour (DPC) => Information about area inside the contour Textural Information (Homogeneity, Uniformity) 79

33 Correct class Automated Clustering Classification technique: unsupervised k-means clustering Test set: 218 vector objects Confusion matrix Prediction C1 C2 C C Classification accuracy Classification accuracy Matthew correlation coefficient Evaluation results 99,5% 0,99 80

34 Priority Research Proposal (DFG) Multisensor Remote Sensing for Landscape Dynamics Assessment (MultiSense): Provision of means for an integrated analysis of multisensor RS datasets for operational change analysis and monitoring within three key regions of global environmental change Development of new algorithms and methodologies for robust geo-biophysical parameter extraction from multisensor RS for landscape dynamics assessment Development and allocation of a powerful Open Source image processing system for multisensor RS analysis

35 Remote Sensing Space Missions

36 Study Sites

37 Thematic Structure

38 Organizational Structure

39 MultiSense About 12 projects total 24 Ph.D. and 4 Postdocs 2 x 3 years Mio /year

40 Conclusions VHR Data: Important/Necessary Fusion Techniques: Iconic to Decision Based RS: From Data Sparse to Data Rich Automated Change Detection: Cooperative + GIS